What should a company ask an AI data buyer before licensing its records?
Short answer. Ask what the records will be used for and who else will see them, whether exclusivity is wanted and for how long, who removes personal information, how delivery is accepted and paid, and what happens to the data and to trained models when the deal ends. The answers show whether the buyer fits before any lawyer time is spent on a draft.
General information, not legal advice.
About the use
Start with the purpose. Will the records train a general model, a model for one industry, an evaluation benchmark, or a training environment that the buyer will sell to others? Each leads to different exposure. If the buyer builds environments for other labs, ask whether those customers receive anything derived from your records.
Ask which parts of the manifest they actually want. A buyer that needs only the help desk history should not ask for the mailbox, and narrowing the scope early makes de-identification and pricing simpler.
About exclusivity and price
Ask whether the offer assumes exclusivity, for how long, and over what: these records only, or the whole category of data from your company. Ask how the price would change without it. Many buyers value a set less if a competitor can buy the same records, so the question is a fair one, not a negotiating trick.
Ask how the figure was reached, whether it is fixed or tied to counts, and when each part is paid. If a buyer asks you to answer quickly, ask what changes after the deadline and get the terms in writing before agreeing.
About privacy and personal information
Ask who removes personal information, what standard they use, and whether you see and approve a sample before anything leaves your systems. Ask who carries the risk if a person turns up in the data later. These answers decide where liability sits.
Ask whether the buyer has promised its own users or customers anything about the data it trains on. The FTC has warned AI companies that they must keep the privacy and confidentiality commitments they make, and a buyer that has made such promises will have firm rules on what it accepts.
About the end of the deal
Ask what happens when the license ends: are raw files deleted, how is deletion confirmed, and may models already trained be kept and used? Ask whether you can audit or receive a certificate. Ask, too, what the buyer does if your company is sold or closes, since the license may need to pass to a new owner.
Questions about the buyer itself
Before the deal terms, learn who you are dealing with. Ask whether the buyer trains models itself or sources data for other labs, since a sourcing company may pass your records on and its own customer sets the real rules. Ask which legal entity would sign, where it is based, and which law would govern the license.
Ask how the buyer evaluates a sample and who on its side makes the decision, so you know whether the person you are talking to can commit. Ask how long its review usually takes and what it needs from you to keep it moving. Ask whether it has licensed records like yours before, and what went wrong if a past deal fell through.
Finally, ask what the buyer will not accept. Many have rules on record types, countries, company size or personal data. Hearing those early saves both sides from preparing a sample that was never going to pass, and it tells you whether to spend more time with this buyer at all.
Follow-up questions
Is it rude to ask a buyer these questions early?
No. Serious buyers expect them and usually have standard answers. A buyer that will not say how the data is used or shared is giving you useful information too.
Should the answers be in writing?
Yes. A short email that confirms use, exclusivity, de-identification and payment terms gives your counsel a fixed point to compare the draft license against.
What if two buyers give different answers?
Compare them side by side on the same points. A higher price with broad reuse rights may be worth less than a lower one limited to a single model.
Didn't find your question? Ask us directly.